********************************************************
BelgiumTSC - The Belgium Traffic Sign Classification Benchmark
********************************************************

This archive contains the testing set of the 
"Belgium Traffic Sign Classification Benchmark".

This testing set is a subset of the BelgiumTS - Belgium Traffic Signs database:
http://homes.esat.kuleuven.be/~rtimofte/
http://www.vision.ee.ethz.ch/~timofter/

and follows the structure used in GTSRB Challenge proposed for IJCNN 2011:
http://benchmark.ini.rub.de/


**********************************************
Archive content
**********************************************
This archive contains the following structure:

There is one directory for each of the 62 classes (00000 - 00061).
Each directory contains the corresponding training images and one 
text file with annotations, eg. GT-00000.csv. 

In total are 2534 images for testing. On average for each physically distinct
traffic sign there are 3 images available.


**********************************************
Image format and naming
**********************************************
The images are PPM images (RGB color). Files are numbered in two parts:

   XXXXX_YYYYY.ppm

The first part, XXXXX, represents the pole number from where is in the full
annotations of BelgiumTS database. All images of one class 
with identical pole numbers originate from the same pole and more represent
one single physical traffic sign.

The second part, YYYYY, is a running number for the views where the traffic
sign is annotated. There is no temporal order
of the images, but they were where extracted in the order of the annotations on
the original pole from BelgiumTS.


**********************************************
Annotation format
**********************************************

The annotations are stored in CSV format (field separator
is ";" (semicolon) ). The annotations contain meta information 
about the image and the class id.


In detail, the annotations provide the following fields:

Filename        - Image file the following information applies to
Width, Height   - Dimensions of the image
Roi.x1,Roi.y1,
Roi.x2,Roi.y2   - Location of the sign within the image
		  (Images contain a border around the actual sign
                  of 10 percent of the sign size, at least 5 pixel)
ClassId         - The class of the traffic sign


**********************************************
How to cite?
**********************************************
@inproceedings{Timofte-BMVC-2011,
author = {Radu Timofte and Luc Van Gool},
title  = {Sparse representation based projections},
booktitle = {British Machine Vision Conference},
year   = {2011},
}

@article{Timofte-MVA-2011,
  author = {Radu Timofte and Karel Zimmermann and Luc {Van Gool}},
  title = {Multi-view Traffic Sign Detection, Recognition, and 3D Localisation},
  journal = {Machine Vision and Applications},
  year = {2011}, 
  doi  = {10.1007/s00138-011-0391-3},
}


**********************************************
Further information
**********************************************
For more information please visit the website at

http://homes.esat.kuleuven.be/~rtimofte/
http://www.vision.ee.ethz.ch/~timofter/

If you have any questions, do not hesitate to contact us 
    
	[email protected]
	[email protected]


**********************************************
Centre for Processing Speech and Images (PSI/ESAT)
VISion in Industry, Communications, and Services (VISICS)

Katholieke Universiteit Leuven
Belgium
**********************************************